Mark interaction as internalizable instead of dropping it

[MartinquaXD]

The optimization pipeline which should improve the objective value of a settlement by applying a few gas optimizations can cause solutions that are not viable to pass the simulation by internalizing AMM interactions that are overly optimistic.

Check out this trade selling USDC for AMPL. Paraswap currently gives too optimistic prices for AMPL. That means the settlement provided by paraswap would likely have reverted. Because the order was selling USDC (a trusted token) the optimization pipeline was allowed to internalize the erroneous AMM interaction which resulted in a working settlement with an unbeatable price (because we pay with our buffers).

I saw 2 ways to tackle this issue:

1. Don't apply optimizations if the original settlement would already revert
2. Instead of right out dropping the interaction we keep the original interaction but mark it as internalizable

I decided to go with 2 because together with #843 it should prevent the same problems while offering better debuggability than 1.

Test Plan

?